This thesis investigates the implementation of a human activity recognition system in smart home contexts, focusing on the use of ambient sensors and sensors on objects. By analyzing sensor data with the use of Deep Learning and Machine Learning techniques, activity recognition systems can discern and interpret human behavior, providing valuable information for a variety of scenarios, including elderly care and health monitoring. The research begins with a comprehensive review of existing datasets, culminating in the selection of a suitable dataset for training activity recognition models. A new metric is introduced to delineate the relationship between activities and sensors, which facilitates a nuanced understanding of predictive ability. Subsequent transformations of the dataset aim to mitigate dependencies on individual residents and enable recognition across diverse user profiles. In addition, methods of encoding temporal information and data augmentation strategies are explored to improve model performance. The thesis concludes with the training and evaluation of ten models, comprising convolutional neural networks and a hybrid architecture combining convolutional neural networks with XGBoost, demonstrating promising capabilities in activity recognition and achieving f-scores of up to 81%.